Merge pull request #203 from PartitionedArrays/release-0.5.12

Release PA 0.5.12 and PS 0.3.3

Author: fverdugo

CHANGELOG.md

@@ -5,6 +5,12 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.5.12] - 2025-06-09
+
+### Fixed
+
+- Miscellaneous performance improvements.
+
 ## [0.5.11] - 2025-06-06
 
 ### Added

PartitionedSolvers/CHANGELOG.md

@@ -6,6 +6,12 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.3.3] - 2025-06-09
+
+### Fixed
+
+- Miscellaneous performance improvements.
+
 ## [0.3.2] - 2025-03-13
 
 ### Added

PartitionedSolvers/Project.toml

@@ -1,7 +1,7 @@
 name = "PartitionedSolvers"
 uuid = "11b65f7f-80ac-401b-9ef2-3db765482d62"
 authors = ["Francesc Verdugo <f.verdugo.rojano@vu.nl>"]
-version = "0.3.2"
+version = "0.3.3"
 
 [deps]
 IterativeSolvers = "42fd0dbc-a981-5370-80f2-aaf504508153"

PartitionedSolvers/src/krylov_solvers.jl

@@ -197,20 +197,22 @@ function cg_update(ws,A)
     update(ws;iteration,Pl,A)
 end
 
-function cg_step(x,ws,b,phase=:start;kwargs...)
+function cg_step(x,ws,b,phase=:start;zero_guess=false,kwargs...)
     (;dx,r,u,A,ρ,iteration,current,target) = ws.state
     (;reltol,abstol,norm,Pl) = ws.options
     s = u
     if phase === :start
         iteration = 0
         phase = :advance
         copyto!(r,b)
-        mul!(s,A,x)
-        axpy!(-one(eltype(s)),s,r)
-        #r .-= s
+        if ! zero_guess
+            mul!(s,A,x)
+            axpy!(-one(eltype(s)),s,r)
+            #r .-= s
+        end
         current = norm(r)
         target = max(reltol*current,abstol)
-        dx .= zero(eltype(dx))
+        fill!(dx,zero(eltype(dx)))
         ρ = one(eltype(x))
         ws = update(ws;iteration,ρ,current,target)
         print_progress_header(ws)
@@ -228,7 +230,6 @@ function cg_step(x,ws,b,phase=:start;kwargs...)
     #x .+= α .* dx
     axpy!(-α,s,r)
     #r .-= α .* s
-    current = norm(r)
     iteration += 1
     current = norm(r)
     ws = update(ws;iteration,ρ,current)

PartitionedSolvers/src/smoothers.jl

@@ -1,17 +1,27 @@
+function identity_solver_step(x,workspace,b,phase=:start;kwargs...)
+    if isa(b,PVector)
+        foreach(copyto!,own_values(x),own_values(b))
+    else
+        copyto!(x,b)
+    end
+    phase = :stop
+    x,workspace,phase
+end
+
+function identity_solver_update(workspace,A)
+    workspace
+end
 
 function identity_solver(p)
     @assert uses_mutable_types(p)
     workspace = nothing
-    function update(workspace,A)
-        workspace
-    end
-    function step(x,workspace,b,phase=:start;kwargs...)
-        copyto!(x,b)
-        phase = :stop
-        x,workspace,phase
-    end
     uses_initial_guess = Val(false)
-    linear_solver(update,step,p,uses_initial_guess)
+    workspace = nothing
+    linear_solver(identity_solver_update,identity_solver_step,p,workspace;uses_initial_guess)
+end
+
+function is_identity_solver(s)
+    s.step == identity_solver_step
 end
 
 function jacobi_correction(p)

Project.toml

@@ -1,7 +1,7 @@
 name = "PartitionedArrays"
 uuid = "5a9dfac6-5c52-46f7-8278-5e2210713be9"
 authors = ["Francesc Verdugo <f.verdugo.rojano@vu.nl> and contributors"]
-version = "0.5.11"
+version = "0.5.12"
 
 [deps]
 Adapt = "79e6a3ab-5dfb-504d-930d-738a2a938a0e"

src/p_sparse_matrix.jl

@@ -2412,14 +2412,15 @@ function LinearAlgebra.mul!(c::PVector,a::PSparseMatrix,b::PVector)
     @boundscheck @assert matching_own_indices(axes(c,1),axes(a,1))
     @boundscheck @assert matching_own_indices(axes(a,2),axes(b,1))
     @boundscheck @assert matching_ghost_indices(axes(a,2),axes(b,1))
-    if ! a.assembled
+    if a.assembled
+        t = consistent!(b)
+        foreach(spmv!,own_values(c),own_own_values(a),own_values(b))
+        wait(t)
+        foreach(muladd!,own_values(c),own_ghost_values(a),ghost_values(b))
+    else
         @boundscheck @assert matching_ghost_indices(axes(a,1),axes(c,1))
-        return mul!(c,a,b,1,0)
+        mul!(c,a,b,one(eltype(a)),zero(eltype(a)))
     end
-    t = consistent!(b)
-    foreach(spmv!,own_values(c),own_own_values(a),own_values(b))
-    wait(t)
-    foreach(muladd!,own_values(c),own_ghost_values(a),ghost_values(b))
     c
 end
 

src/p_vector.jl

@@ -1540,13 +1540,44 @@ function renumber(a::PVector,row_partition_2;renumber_local_indices=Val(true))
     PVector(values,row_partition_2)
 end
 
+
 function LinearAlgebra.axpy!(a::Number,x::PVector,y::PVector)
-    y .+= a .* x
+    #y .+= a .* x
+    function axpy_local!(x,y)
+        ni = length(y)
+        @boundscheck @assert length(x) == ni
+        for i in 1:ni
+            @inbounds y[i] += a * x[i]
+        end
+    end
+    function axpy_local!(x::Vector,y::Vector)
+        ni = length(y)
+        @boundscheck @assert length(x) == ni
+        @simd for i in 1:ni
+            @inbounds y[i] += a * x[i]
+        end
+    end
+    foreach(axpy_local!,own_values(x),own_values(y))
     y
 end
 
 function LinearAlgebra.axpby!(a::Number,x::PVector,b::Number,y::PVector)
-    y .= a .* x .+ b .* y
+    #y .= a .* x .+ b .* y
+    function axpby_local!(x,y)
+        ni = length(y)
+        @boundscheck @assert length(x) == ni
+        for i in 1:ni
+            @inbounds y[i] = a * x[i] + b * y[i]
+        end
+    end
+    function axpby_local!(x::Vector,y::Vector)
+        ni = length(y)
+        @boundscheck @assert length(x) == ni
+        @simd for i in 1:ni
+            @inbounds y[i] = a * x[i] + b * y[i]
+        end
+    end
+    foreach(axpby_local!,own_values(x),own_values(y))
     y
 end